Generative AI

GenAI

An overview

Author

Chi Zhang

Published

October 29, 2024

Generative AI

AI systems that can produce high quality content, specifically text, images and audio

General purpose technology: like electricity, and AI

  • web based: chat-gpt, bard, …
  • software based

History of LLM

  • ELiza (1960s), pattern matching. Encourages introspection, creating the illusion of understanding to make people feel heard
  • 2010s, GPT - generative pre-trained transformer by OpenAI
    • GPT2 (2018)
    • GPT3 (2020)
    • GPT4
  • LlaMA by Meta

Prompt engineering

Write, refine, optimize prompts; perfect the interaction between humans and AI; continuously monitor these prompts; maintain an up-to-date prompt library

Prompting guide

  • give clear, specific prompt with context
  • ask to include sources
  • avoid leading the answer such that the answer is biased

Quality of the results

  • accuracy
  • unbiased
  • sufficient information
  • consistency if using the same prompt multiple times

Shot (examples): zero shot, one shot, few shot prompting (need to provide follow up information to make the answer more customized)

Chain of thought prompting: break down a large task into smaller sub-tasks